Reinforcement learning is the study of methods for learning how to act in environments in order to maximize reward. One of the key ideas in reinforcement learning is the concept of a state-action value function, or $Q$-function. In this talk we will precisely define the reinforcement learning problem, develop the idea of $Q$-learning to solve it, and extend $Q$-learning with deep neural networks to give deep $Q$-networks. As a result, we'll see neural networks that can learn on their own how to balance a pole on a cart, safely land a spaceship, or even play Atari games far better than humans.